BLADE: Auditing Information Use for Failure Attribution in LLM Multi-Agent Systems
We introduce BLADE (Blame Localization via Audited Dependency Evidence), a training-free framework that moves beyond tracing information to auditing its use.
ProofPaper ↗
Key points
- Large language model multi-agent systems extend individual agents' capabilities through collaboration across specialized roles, tools, and workflows.
- Yet collaboration makes failures harder to diagnose: one agent may misinterpret a correct result while others carry the mistake forward.
- Using Jev for structured probabilistic judgments, BLADE uses dependency probabilities to select relevant source-decision handoffs for auditing.
- Making responsibility evidence explicit offers a path toward more interpretable and reliable agentic systems.
Sources (1)
- [1]BLADE: Auditing Information Use for Failure Attribution in LLM Multi-Agent SystemsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 05:34 PM
We introduce BLADE (Blame Localization via Audited Dependency Evidence), a training-free framework that moves beyond tracing information to auditing its use.
Large language model multi-agent systems extend individual agents' capabilities through collaboration across specialized roles, tools, and workflows.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026Recalibrating Many Questions with Few Labels: Minimax Rates and Identification Limits
- Oct 13, 2026Judges Follow the Reference: The Derivation Gap in Verifying Tool-Call Arguments
- Oct 12, 2026Releasing the Inner Jev: A Simple Recipe for Open Decision Models
- Oct 12, 2026Deploy decision models with llama.cpp on Amazon SageMaker AI
- Oct 12, 2026Correct Choices, Failed Sequences: Choice2Seq Benchmarks Correctness Retention Across Output Interfaces
- Oct 12, 2026Show HN: Atticus – Hold-to-talk voice assistant for macOS that runs on Claude -p